CN108090507A - A kind of medical imaging textural characteristics processing method based on integrated approach - Google Patents
A kind of medical imaging textural characteristics processing method based on integrated approach Download PDFInfo
- Publication number
- CN108090507A CN108090507A CN201711320261.1A CN201711320261A CN108090507A CN 108090507 A CN108090507 A CN 108090507A CN 201711320261 A CN201711320261 A CN 201711320261A CN 108090507 A CN108090507 A CN 108090507A
- Authority
- CN
- China
- Prior art keywords
- textural characteristics
- feature
- training set
- medical imaging
- tree
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/24323—Tree-organised classifiers
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/40—Analysis of texture
- G06T7/41—Analysis of texture based on statistical description of texture
- G06T7/45—Analysis of texture based on statistical description of texture using co-occurrence matrix computation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20092—Interactive image processing based on input by user
- G06T2207/20104—Interactive definition of region of interest [ROI]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30096—Tumor; Lesion
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Multimedia (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Radiology & Medical Imaging (AREA)
- Quality & Reliability (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Probability & Statistics with Applications (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of medical imaging textural characteristics processing methods based on integrated approach, are related to medical digital images processing technology field, and the present invention includes herein below：With the integrated approach based on tree in machine learning, the textural characteristics extracted using gray level co-occurrence matrixes and Curvelet conversion from medical imaging are modeled, to predict target variable, the method of the present invention to feature by first carrying out dimensionality reduction, then model again, the non-linear relation between feature can be efficiently used, the miscellaneous function of rational judgment is provided, auxiliary doctor judges medical imaging.
Description
Technical field
The present invention relates to medical digital images processing technology field, more particularly to a kind of doctor based on integrated approach
Treat image texture characteristic processing method.
Background technology
Medical imaging is all important information source and assistance tool of the doctor when carrying out the work all the time, in clinical work
Medical image energy provides more rich, more accurate diagnosis and treatment information for doctor in work and scientific research.With the development of information age, especially
It is the rise of big data, artificial intelligence field, excavating the more useful informations hidden in medical imaging for depth provides effectively
Instrument, from image the variation of pixel find out the interested information of doctor.
The textural characteristics of medical imaging are primarily referred to as the local feature of image, are recurrent local modes in the picture
And its queueing discipline, the textural characteristics of a certain position and the gray-value variation rule around the position are closely related in image；Together
When, textural characteristics are also a kind of global characteristics, can be used for describing image or the surface nature corresponding to image-region.Texture point
Analysis refers to extract textural characteristics by certain image processing techniques, so as to obtain the processing procedure of the quantitative description of texture.
Certain variation of image pixel gray level grade or color can be described by texture analysis, image texture characteristic and structure can be done
Go out quantitative description and explanation.It is formed since texture is occurred repeatedly by intensity profile on spatial position, thus in image
Being separated by space between two pixels of certain distance can be there are certain gray-scale relation, i.e., the space correlation feature of gray scale in image.
Gray level co-occurrence matrixes (GLCM) are exactly a kind of to analyze the important of image texture characteristic by studying the spatial correlation characteristic of gray scale
Method.
One medical imaging can construct gray level co-occurrence matrixes respectively on 12 directions, this 12 directions and time
Decoupled direction it is consistent, each gray level co-occurrence matrixes pass through multi-scale geometric analysis, be usually continuous Qu Bo (Curvelet)
Conversion, can obtain 16 textural characteristics, they are respectively：
Traditional statistical method, such as variance analysis, similarity calculation, because it can not calculate multiple features and target
Correlation between variable, so respective 16 textural characteristics on this 12 directions can not be handled well.
The content of the invention
It is an object of the invention to：In order to solve existing statistical method can not calculate multiple features and target variable it
Between correlation the problem of, the present invention provides a kind of medical imaging textural characteristics processing method based on integrated approach, with machine
The integrated approach based on tree in device study, to what is extracted using gray level co-occurrence matrixes and continuous warp wavelet from medical imaging
Textural characteristics are modeled, and carry out dimensionality reduction extraction to textural characteristics before modeling, are efficiently used non-linear between feature
Relation, provides the miscellaneous function of rational judgment, and auxiliary doctor judges medical imaging.
The present invention specifically uses following technical scheme to achieve these goals：
A kind of medical imaging textural characteristics processing method based on integrated approach, comprises the following steps：
Step 1: texture feature extraction, using gray level co-occurrence matrixes and continuous warp wavelet from m sample of medical imaging
In extract k textural characteristics K (K1,...,Kk), using the matrix of one m × k of m sample and k textural characteristics structure, and
Matrix cutting is collected for training set and verification according to a certain percentage, and the quantity of training intensive data is more than verification and collects；
Step 2: feature normalization, allow between textural characteristics and transform to same size, the texture for avoiding excursion big is special
Sign occupies more weights, causes Biased estimator, can accelerate modeling speed, evades the not convergent risk of model, is specifically：To instruction
Practice the textural characteristics concentrated and carry out feature normalization processing, form new training set, and same standard coefficient is applied to
On verification collection, feature normalization processing is carried out to the textural characteristics that verification is concentrated, forms new verification collection；
Step 3: Feature Dimension Reduction, with tree-model to the textural characteristics dimensionality reduction in new training set, is specifically：According to
The number that textural characteristics are selected in tree-model carries out ranking to textural characteristics from more to less, n textural characteristics before extracting, remaining
Under training set then be dimensionality reduction training set；
Step 4: modeling, carries out machine learning method modeling with random forest to dimensionality reduction training set, obtains random forest
Model；
Step 5: output as a result, predicted with Random Forest model new verification collection, exports the pre- of new verification collection
Classification is surveyed to get to prediction result.
In above-mentioned technical proposal, the calculation formula that feature normalization is handled in the step 2 is：In formula S is the set of textural characteristics in training set, and i is the number of textural characteristics in training set.
In above-mentioned technical proposal, the basic mode type of tree-model uses post-class processing (CART), step 4 in the step 3
In further include the quantity to basic mode type, the depth of tree and the number of leaf node and be configured.
CART, using top-to-bottom method, selects the feature of one " best " " to divide in building process in each step
Split " tree, and the definition of " best " is so that the training dataset in child node is tried one's best is pure, different algorithms use different fingers
It marks to define " best ", CART is then using GINI indexes, and node is more impure, and GINI values are bigger, and carries out tree in selected characteristic
The GINI values of leaf node will be smaller than the GINI values of father node after division, and different features corresponds to different decrements,
CART chooses the partitioning standards of the feature present node of decrement maximum.In tree-model, the number that each feature is selected is got over
It is more, indicate that this feature is more important, the number being selected according to each feature can carry out the significance level of each feature
Ranking, n feature carries out next step modeling before selection.
In above-mentioned technical proposal, in the step 3,5%~20% of sample size in training set new n=.
In above-mentioned technical proposal, in the step 1, the cutting ratio of matrix is 7:3.
Integrated approach is one model of structure, and the predicted value of base grader is combined with certain means, integrates base grader
Model generalization ability and robustness afterwards is better than the model constructed by single base grader.It can by the difference integrated in means
Integrated approach is divided into as Bagging and Boosting methods, Bagging methods are in the random subset of original training data collection
Base grader is built, then polymerize its respective prediction to form final prediction, Bagging methods will be by that will be randomized introducing
Its building process and then the mode polymerizeing reduce the variance of grader.Since this method can efficiently reduce the wind of over-fitting
Danger, the base grader of Bagging methods can use strong classifier (for example, decision tree of fully nonlinear water wave), this is with using weak point
Class device (for example, shallow-layer decision tree) usually most effective Boosting methods are on the contrary, in contrast, in Boosting methods,
Base grader is built in order, and attempts to reduce the deviation of block mold.
For the present invention using machine learning classification algorithm of the decision tree as the base grader of integrated approach, decision tree is a kind of
Discriminative model, in a categorised decision tree, non-leaf nodes is decision rule, and leaf node is classification.When input one
During feature vector, moved according to the rule on decision tree from root node to leaf node, it is finally defeated according to the kind judging of leaf node
The classification of incoming vector.
Establishing a decision-tree model, mainly there are three steps：Feature selecting, the generation of decision tree and the beta pruning of decision tree.
Information gain (Information Gain) is used during feature selecting, by name come if understanding, be exactly front and rear information difference
Value in decision tree classification problem, is exactly information difference of the decision tree before Attributions selection division is carried out and after division, calculates public
Formula is：
G (D, A)=H (D)-H (D | A) (1)
In formula, D is the stochastic variable of data category, and A is the stochastic variable of some feature of data, and information gain is known
Probabilistic reduction degree of stochastic variable D in the case of A, that is, it is understood that how many letter obtained in the case of A
Breath, it can be seen that so that the feature of information gain maximum is exactly best feature, because this feature can be maximum from (1) formula
Our uncertainties to classification are reduced in degree, so in decision-tree model, use information gain carries out each layer
Feature selecting.
The difference in the calculation of information is included according to decision tree, different trees can be marked off, the present invention uses
Base grader for CART (Classification And Regression Tree, post-class processing), informational content measure
Mode is GINI indexes, it is assumed that data have n classification, the calculation formula of that GINI is：
In formula, piRepresent that sample belongs to the probability of i classes.
After the scheme of feature selecting determines, decision tree is established by following steps：
Root node is set to all categories by step 1；
Step 2, selection make the feature of information gain maximum divide root node, generate several child nodes；
Step 3, recurrence carry out above-mentioned steps.
End condition is set during decision tree is established, and the condition of typically ending up is that the data in some node belong to
Same class, the end condition also having are that information gain is less than some threshold value etc..When the data category in leaf node is inconsistent, make
The mode of classification is as the classification of the leaf node and the result of empirical risk minimization in by the use of the node.
If the level of decision tree is very deep, there are many leaf node, then can have good classification results to training data.
The depth of decision tree is bigger or the leaf node of decision tree is more, and model is also more complicated, easier generation over-fitting, at this moment
It needs to carry out regularization, that is, the model complexity of decision tree is punished.
The leaf node number of decision tree is T, and t is the leaf node of decision tree, which has NtA sample, wherein belonging to n-th
Class has NtnIt is a, Ht(T) it is the GINI values on leaf node t, the calculation formula of the study loss function of decision tree CART is defined as：
In formula,Model is represented to the error of training data, α | T | represent the complexity of model, α is to miss
Difference and the balance factor of complexity between the two can train the CART of Structural risk minization according to formula (2).
Although by can effectively avoid over-fitting problem by adding in regular terms in CART loss functions, due to
CART is simple Weak Classifier, and there is the noise of many separate sources in the textural characteristics of medical imaging, and count
It is strong according to isomerism, cause just at last an optimal CART can not also be fitted real data well, one is effective
Method is Bagging+CART to be modeled to data, this method is random forest (Random Forest).
Random forest is exactly to establish a forest with random manner, is made of inside forest many CART, each
It is no associated between CART, after forest establishes, as soon as when new input sample, allows each in forest
CART respectively predicts its classification, judges the final prediction classification of new samples according to HUTD Algorithm's.
The core of random forest is " random ", i.e. randomness in sample sampling and feature sampling.Each base is classified
The input sample of device is to be sampled by original sample in a manner of putting back to, selection when node split is carried out of base grader
The subset of characteristic set is calculated, and due to the introducing of the two randomnesss, reduces the correlation of each base grader, table
Generally it is exactly the ability for improving the anti-over-fitting of model now.
The outstanding place of random forest is also embodied in：Although black-box model, it is explanatory to the significance level of feature
By force, training speed is fast, may be readily formed as parallel method, when creating random forest, to extensive error using unbiased esti-mator
Deng.
Beneficial effects of the present invention are as follows：
1st, due to medical image data gather it is difficult, and doctor when area-of-interest is delineated due to experienced
Difference can cause error, this results in textural characteristics (feature) number much larger than the number of pathology (sample), and data can have mistake
Difference, the present invention before machine learning model is established to data the step of introduced feature dimensionality reduction, and the method for dimensionality reduction in order to build
Being consistent property of mould method, the method based on random forest that used random tree this reduce error by dimensionality reduction, effectively profit
With the non-linear relation between feature, so as to carry out more accurately prediction to target variable, auxiliary doctor to medical imaging into
Row judges.
2nd, for the training dataset of each base grader, random forest is sampled using bootstrap, is used as
The training dataset of each decision tree, and random tree makees base grader training using raw data set, is having selected division
After feature, the base grader of random forest can be based on information gain, and Gini coefficient method selects an optimal feature value division
Point, but random tree strengthens randomness, can be randomly chosen a characteristic value to divide decision tree, due to having randomly choosed feature
The division points position of value rather than optimum point position can so cause the scale of the decision tree of generation to be generally larger than random forest and be given birth to
Into decision tree, that is to say, that the variance of model can be reduced compared with random forest, but deviation can increase compared with random forest
Greatly, deviation increase just can learn the potential relation between feature and classification well, and since it is with important to feature
Degree it is explanatory, we just can extract effective textural characteristics with random tree to achieve the purpose that Feature Dimension Reduction, provide
The miscellaneous function of rational judgment, auxiliary doctor judge medical imaging.
Description of the drawings
Fig. 1 is the general frame flow diagram of the present invention.
Fig. 2 is the relation schematic diagram between feature before dimensionality reduction of the present invention.
Fig. 3 is the relation schematic diagram between feature after dimensionality reduction of the present invention
Fig. 4 is workflow schematic diagram of the random forest of the present invention in the training pattern stage.
Specific embodiment
In order to which those skilled in the art are better understood from the present invention, below in conjunction with the accompanying drawings with following embodiment to the present invention
It is described in further detail.
Embodiment 1
As shown in Figures 1 to 4, the present embodiment proposes a kind of medical imaging textural characteristics processing side based on integrated approach
Method comprises the following steps：
Step 1: texture feature extraction, using gray level co-occurrence matrixes and continuous warp wavelet from m sample of medical imaging
In extract k textural characteristics K (K1,...,Kk), using the matrix of one m × k of m sample and k textural characteristics structure, and
Matrix cutting is collected for training set and verification according to a certain percentage, and the quantity of training intensive data is more than verification and collects；
Step 2: the textural characteristics in training set are carried out feature normalization processing, form new training by feature normalization
Collection, and same standard coefficient is applied on verification collection, feature normalization processing is carried out to the textural characteristics that verification is concentrated,
Form new verification collection；
Step 3: Feature Dimension Reduction, with tree-model to the textural characteristics dimensionality reduction in new training set, is specifically：According to
The number that textural characteristics are selected in tree-model carries out ranking to textural characteristics from more to less, n textural characteristics before extracting, remaining
Under training set then be dimensionality reduction training set；
Step 4: modeling, carries out machine learning method modeling with random forest to dimensionality reduction training set, obtains random forest
Model；
Step 5: output as a result, predicted with Random Forest model new verification collection, exports the pre- of new verification collection
Classification is surveyed to get to prediction result.
In above-mentioned technical proposal, the calculation formula that feature normalization is handled in the step 2 is：In formula S is the set of textural characteristics in training set, and i is the number of textural characteristics in training set.
In above-mentioned technical proposal, the basic mode type of tree-model uses post-class processing (CART), step 4 in the step 3
In further include the quantity to basic mode type, the depth of tree and the number of leaf node and be configured.
CART, using top-to-bottom method, selects the feature of one " best " " to divide in building process in each step
Split " tree, and the definition of " best " is so that the training dataset in child node is tried one's best is pure, different algorithms use different fingers
It marks to define " best ", CART is then using GINI indexes, and node is more impure, and GINI values are bigger, and carries out tree in selected characteristic
The GINI values of leaf node will be smaller than the GINI values of father node after division, and different features corresponds to different decrements,
CART chooses the partitioning standards of the feature present node of decrement maximum.In tree-model, the number that each feature is selected is got over
It is more, indicate that this feature is more important, the number being selected according to each feature can carry out the significance level of each feature
Ranking, n feature carries out next step modeling before selection.
Due to medical image data gather it is difficult, and doctor when area-of-interest is delineated due to experienced difference
Different to cause error, this results in textural characteristics number much larger than the number of pathology, and data can be there are error, and the present embodiment is right
Data are established the step of introduced feature dimensionality reduction before machine learning model, and the method for dimensionality reduction with modeling method in order to be consistent
Property, the method based on random forest that used random tree this reduce error by dimensionality reduction, efficiently use non-between feature
Linear relationship judges medical imaging so as to carry out more accurately prediction, auxiliary doctor to target variable.
Embodiment 2
Based on embodiment 1, the present embodiment proposes a kind of medical imaging textural characteristics processing method based on integrated approach,
Specifically include following steps：
Step 1: doctor takes medical imaging, interested focal part in medical imaging is sketched out, according to focal part
Sample, the gray level co-occurrence matrixes of calculating foci part, to gray level co-occurrence matrixes with continuous warp wavelet generate textural characteristics, profit
With textural characteristics and target variable (such as tumour grade classification, prognostic judgement etc.) one matrix of structure；
Step 2: matrix according to 7:3 ratio cutting collects for training set and verification, if target variable is unbalanced class
Not, to be solved in cutting training set and verification collection by way of stratified sampling；
Step 3: carrying out feature normalization processing to the feature of training set, new training set is obtained, and same standard
Change coefficient to apply on verification collection, obtain new verification collection；
Step 4: with random tree come training data on new training set, according in random tree textural characteristics be selected
Number from more to less to textural characteristics carry out ranking, the feature that n textural characteristics are inputted as machine learning model before selection
The 5%~20% of sample size () in training set new n=；
It is built Step 5: being trained with random forest having chosen remaining training set after textural characteristics to step 4
Mould, random forest have a important parameter, that is, the quantity k of random forest lining tree, k for odd number and k=max (201, it is remaining
Training set in sample size 50%)；
Step 6: new verification collection is input in step 5 in established model, the pre- of new verification collection is exported
Classification is surveyed, with general evaluation index (such as accuracy rate, ROC (Receiver Operating Characteristic, recipient
Operating characteristics) curve etc.) come the generalization abilities of judgment models, that is, predictive ability of the model on unknown data collection.
Medical image since data acquisition is difficult, and doctor delineate ROI (region of interest, it is interested
Region) when error can be caused due to experienced difference, this results in textural characteristics (feature) number much larger than pathology (sample
Originally number), and data can be there are error, the step of present invention introduced feature dimensionality reduction before machine learning model is established to data
Suddenly, and the method for dimensionality reduction is in order to being consistent property of modeling method, the method based on random forest that used random tree this, this
Kind of method and random forest difference lies in：For the training dataset of each base grader, random forest using
Bootstrap is sampled, as the training dataset of each decision tree, and random tree using raw data set come to base grader
Do training.After division feature is had selected, the base grader of random forest can be based on information gain, Gini coefficient method, selection
One optimal feature value division point, but random tree strengthens randomness, can be randomly chosen a characteristic value to divide decision-making
Tree due to having randomly choosed the division points position of characteristic value rather than optimum point position, can so cause the scale of the decision tree of generation
The decision tree that generally larger than random forest is generated, that is to say, that the variance of model can be reduced compared with random forest, but partially
Difference can increase compared with random forest, and deviation increase just can learn the potential relation between feature and classification well, and
Since it has to the explanatory of feature significance level, we just can extract effective textural characteristics with random tree to reach
The purpose of Feature Dimension Reduction, the miscellaneous function of rational judgment is provided for doctor, and auxiliary doctor judges medical imaging.
The above is only presently preferred embodiments of the present invention, is not intended to limit the invention, patent protection model of the invention
It encloses and is subject to claims, the equivalent structure variation that every specification and accompanying drawing content with the present invention is made, similarly
It should include within the scope of the present invention.
Claims (5)
1. a kind of medical imaging textural characteristics processing method based on integrated approach, which is characterized in that comprise the following steps：
Step 1: texture feature extraction, is carried using gray level co-occurrence matrixes and continuous warp wavelet from m sample of medical imaging
Take out k textural characteristics K (K1,...,Kk), utilize m sample and k textural characteristics K (K1,...,Kk) one m × k of structure
Matrix, and matrix cutting is collected for training set and verification, and the quantity of training intensive data is more than verification and collects；
Step 2: the textural characteristics in training set are carried out feature normalization processing, form new training set by feature normalization,
And apply to same standard coefficient on verification collection, feature normalization processing, shape are carried out to the textural characteristics that verification is concentrated
The verification collection of Cheng Xin；
Step 3: Feature Dimension Reduction, with tree-model to the textural characteristics dimensionality reduction in new training set, is specifically：According in tree mould
The number that textural characteristics are selected in type carries out ranking to textural characteristics from more to less, n textural characteristics before extracting, remaining
Training set is dimensionality reduction training set；
Step 4: modeling, carries out machine learning method modeling with random forest to dimensionality reduction training set, obtains random forest mould
Type；
Step 5: output as a result, predicted with Random Forest model new verification collection, exports the prediction class of new verification collection
Not to get to prediction result.
2. a kind of medical imaging textural characteristics processing method based on integrated approach according to claim 1, feature exist
In：The calculation formula of feature normalization processing is in the step 2：In formula
S is the set of textural characteristics in training set, and i is the number of textural characteristics in training set.
3. a kind of medical imaging textural characteristics processing method based on integrated approach according to claim 1 or 2, feature
It is：The basic mode type of tree-model uses post-class processing in the step 3, and the quantity to basic mode type, tree are further included in step 4
Depth and the number of leaf node be configured.
4. a kind of medical imaging textural characteristics processing method based on integrated approach according to claim 1, feature exist
In：In the step 3,5%~20% of sample size in training set new n=.
5. a kind of medical imaging textural characteristics processing method based on integrated approach according to claim 1, feature exist
In：In the step 1, the cutting ratio of matrix is 7:3.
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2017109803837 | 2017-10-19 | ||
| CN201710980383 | 2017-10-19 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108090507A true CN108090507A (en) | 2018-05-29 |
Family
ID=62175336
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711320261.1A Pending CN108090507A (en) | 2017-10-19 | 2017-12-12 | A kind of medical imaging textural characteristics processing method based on integrated approach |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108090507A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110464345A (en) * | 2019-08-22 | 2019-11-19 | 北京航空航天大学 | A kind of separate head bioelectrical power signal interference elimination method and system |
| CN110751201A (en) * | 2019-10-16 | 2020-02-04 | 电子科技大学 | SAR equipment task failure cause reasoning method based on textural feature transformation |
| CN111062442A (en) * | 2019-12-20 | 2020-04-24 | 支付宝(杭州)信息技术有限公司 | Method and device for explaining service processing result of service processing model |
| CN113257000A (en) * | 2021-02-19 | 2021-08-13 | 中用科技有限公司 | Intelligent detection early warning system and method for road black ice |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103699904A (en) * | 2013-12-25 | 2014-04-02 | 大连理工大学 | Image computer-aided diagnosis method for multi-sequence nuclear magnetic resonance images |
| CN105069315A (en) * | 2015-08-26 | 2015-11-18 | 中国环境科学研究院 | Metal shape and validity based aquatic toxicity prediction method |
| CN105931224A (en) * | 2016-04-14 | 2016-09-07 | 浙江大学 | Pathology identification method for routine scan CT image of liver based on random forests |
| CN106815481A (en) * | 2017-01-19 | 2017-06-09 | 中国科学院深圳先进技术研究院 | A kind of life cycle Forecasting Methodology and device based on image group |
| CN107194937A (en) * | 2017-05-27 | 2017-09-22 | 厦门大学 | Tongue image partition method under a kind of open environment |
-
2017
- 2017-12-12 CN CN201711320261.1A patent/CN108090507A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103699904A (en) * | 2013-12-25 | 2014-04-02 | 大连理工大学 | Image computer-aided diagnosis method for multi-sequence nuclear magnetic resonance images |
| CN105069315A (en) * | 2015-08-26 | 2015-11-18 | 中国环境科学研究院 | Metal shape and validity based aquatic toxicity prediction method |
| CN105931224A (en) * | 2016-04-14 | 2016-09-07 | 浙江大学 | Pathology identification method for routine scan CT image of liver based on random forests |
| CN106815481A (en) * | 2017-01-19 | 2017-06-09 | 中国科学院深圳先进技术研究院 | A kind of life cycle Forecasting Methodology and device based on image group |
| CN107194937A (en) * | 2017-05-27 | 2017-09-22 | 厦门大学 | Tongue image partition method under a kind of open environment |
Non-Patent Citations (2)
| Title |
|---|
| YUANJIE ZHENG 等: "Landmark matching based retinal image alignment by enforcing sparsity in correspondence matrix", 《ELSEVIER: MEDICAL IMAGE ANALYSIS》 * |
| 曹生才: "基于内容的医学图像检索技术研究与应用", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110464345A (en) * | 2019-08-22 | 2019-11-19 | 北京航空航天大学 | A kind of separate head bioelectrical power signal interference elimination method and system |
| CN110751201A (en) * | 2019-10-16 | 2020-02-04 | 电子科技大学 | SAR equipment task failure cause reasoning method based on textural feature transformation |
| CN110751201B (en) * | 2019-10-16 | 2022-03-25 | 电子科技大学 | SAR equipment task failure cause reasoning method based on textural feature transformation |
| CN111062442A (en) * | 2019-12-20 | 2020-04-24 | 支付宝(杭州)信息技术有限公司 | Method and device for explaining service processing result of service processing model |
| CN111062442B (en) * | 2019-12-20 | 2022-04-12 | 支付宝(杭州)信息技术有限公司 | Method and device for explaining service processing result of service processing model |
| CN113257000A (en) * | 2021-02-19 | 2021-08-13 | 中用科技有限公司 | Intelligent detection early warning system and method for road black ice |
| CN113257000B (en) * | 2021-02-19 | 2022-10-25 | 中用科技有限公司 | Intelligent detection early warning system and method for road black ice |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109493308B (en) | Medical image synthesis and classification method for generating confrontation network based on condition multi-discrimination | |
| CN112150478B (en) | Method and system for constructing semi-supervised image segmentation framework | |
| CN109948693B (en) | Hyperspectral image classification method based on superpixel sample expansion and generation countermeasure network | |
| CN111191660B (en) | Colon cancer pathology image classification method based on multi-channel collaborative capsule network | |
| CN107977671A (en) | A kind of tongue picture sorting technique based on multitask convolutional neural networks | |
| CN108257135A (en) | The assistant diagnosis system of medical image features is understood based on deep learning method | |
| CN108090507A (en) | A kind of medical imaging textural characteristics processing method based on integrated approach | |
| CN106296699A (en) | Cerebral tumor dividing method based on deep neural network and multi-modal MRI image | |
| CN105407819A (en) | Apparatus, method, and system for automated, non-invasive cell activity tracking | |
| CN110782427B (en) | Magnetic resonance brain tumor automatic segmentation method based on separable cavity convolution | |
| CN111325750B (en) | Medical image segmentation method based on multi-scale fusion U-shaped chain neural network | |
| CN109447998A (en) | Based on the automatic division method under PCANet deep learning model | |
| CN114693933A (en) | Medical image segmentation device based on generation of confrontation network and multi-scale feature fusion | |
| CN117036386A (en) | Cervical MRI image self-supervision segmentation method for generating data by using diffusion model | |
| CN102945553A (en) | Remote sensing image partition method based on automatic difference clustering algorithm | |
| CN115496720A (en) | Gastrointestinal cancer pathological image segmentation method based on ViT mechanism model and related equipment | |
| CN115661165A (en) | Glioma fusion segmentation system and method based on attention enhancement coding and decoding network | |
| Cao et al. | 3D convolutional neural networks fusion model for lung nodule detection onclinical CT scans | |
| CN116863341B (en) | Crop classification and identification method and system based on time sequence satellite remote sensing image | |
| CN117522891A (en) | 3D medical image segmentation system and method | |
| CN116129193B (en) | Method, system and equipment for predicting organoid growth | |
| Li et al. | A deeply supervised convolutional neural network for brain tumor segmentation | |
| CN109934352A (en) | The automatic evolvement method of model of mind | |
| CN116228759A (en) | Computer-aided diagnosis system and apparatus for renal cell carcinoma type | |
| CN114565762B (en) | Weak supervision liver tumor segmentation method based on ROI and split fusion strategy |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180529 |
|
| RJ01 | Rejection of invention patent application after publication |